text-generation-inference

Fixing rocm gptq by using triton code too (renamed cuda into triton).

#2691

Merged